Model card

Reactor 3D

Reduced, interactive plasma-state and alpha-trajectory simulation — not a first-principles reactor prediction. A weighted-macro-particle model of a p-¹¹B mirror device, run in the browser via the real reactor3d_core kernel. The plasma state is a static reduced background (density and temperature are fixed during a run); alpha trajectories are time-stepped. It is not yet a time-evolving fluid twin.

No net-energy, reactor-performance or engineering-feasibility claim. Fidelity is reduced and labelled. Geometry is an abstract confinement box — never a fabrication drawing.

Model levels (they differ)

  • 3D-L0 — ideal upper bound: loss-free direct-conversion collector, no collisional stopping. An analytical ceiling.
  • 3D-L1 — loss-aware (default): collisional stopping/deposition, detector efficiency, parameterized DirectConvert losses, numerical warnings.
  • 3D-L2 — trajectory/field-map study (institutional, future).
  • L0 and L1 produce distinct run identities; L0 recovers ≥ L1 in the collector channel.

Physics

  • Per-cell fluid densities/temperatures with a radial profile (static within a run)
  • Fusion rate R = n_p n_B ⟨σv⟩ (illustrative reactivity proxy)
  • Poisson fusion events → 3 alphas (simplified or distributed energies)
  • Boris pusher in analytic E/B fields; L1 stopping/deposition
  • Modular detectors; DirectConvert collector with fixed configured stages

Equations (resolvable)

Two separate energy ledgers

  • Alpha birth-energy budget (closes to residual): source = deposited + detector + DC-recovered + DC-loss + escaped + residual.
  • Plasma bremsstrahlung is a separate diagnostic, not a sink inside the alpha budget.
  • The viewer labels the alpha table and shows bremsstrahlung distinctly.

DirectConvert coupling

  • Uses the same configured stage potentials for every alpha (not a per-particle optimum).
  • Stages, acceptance and loss versions come from the RunSpec direct_conversion block.
  • L0 uses the ideal (loss-free) collector; L1 uses the seven-channel loss model.

Visualization provenance

  • Plasma points are sampled from the real density field, seeded by run identity — deterministic, not Math.random().
  • Field lines, detectors and the DirectConvert region come from the returned visualization payload.
  • Every alpha has a unique particle_id; trails follow individual particles.

Numerical guards

  • Strict finite conversion of every input (blank/NaN/string rejected with a clear message)
  • float64 density/temperature guards; NaN/Inf & shape trajectory detection
  • Alpha-energy residual tolerance (run flagged FAILED if exceeded)

Verification

  • Alpha-energy conservation to machine precision (~1e-14)
  • Boris gyroradius r = mv/(qB) in uniform B
  • Seed determinism → reproducible run identity (live == sample == package)
  • L0≠L1, unique particle IDs, fixed DC stages (reactor tests)

Unsupported

  • Time-evolving plasma (density/temperature feedback, fuel depletion)
  • Full plasma transport / turbulence
  • Real coil/electrode/vacuum engineering; validated performance or net-energy

Status

  • External validation: Pending (WarpX/Geant4 truth runs, review)
  • Version: reactor3d_core 0.1.0
  • Reviewer state: unreviewed